---
id: 20260801-T0-09
title: "超越KV重建：为MLA草稿模型引入功能重建加速解码"
title_en: "Functional Reconstruction Boosts MLA Draft Models in Speculative Decoding"
url: https://ai.daily.yangsir.net/daily/20260801-T0-09
issue_date: 2026-08-01
publish_date: 2026-07-31T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2607.27269
---

# 超越KV重建：为MLA草稿模型引入功能重建加速解码

arXiv新研究提出针对MLA（多头潜在注意力）架构草稿模型的功能重建方案，区别于传统KV缓存重建。该方法在推测解码中重建功能的潜在状态而非KV本身，大幅减少内存访问，提升长上下文推理吞吐量。实验显示，对解码内存受限的开放权重模型，该方法可提高解码速度和批处理效率，便于长文档处理。

## English Version

**Functional Reconstruction Boosts MLA Draft Models in Speculative Decoding**

A new arXiv paper introduces functional reconstruction for MLA-based draft models in speculative decoding. Instead of rebuilding KV caches, the method reconstructs functional latent states, cutting memory traffic during inference. Experiments show higher decoding throughput and batch efficiency for memory-bound open-weight models—useful for long-context tasks on limited hardware.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2607.27269)

**详情页**：https://ai.daily.yangsir.net/daily/20260801-T0-09

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*